Kelly Criterion for MLB Betting: Bankroll Maths for UK Punters

Editorial Team «mlb Betting Systems» junio 9, 2026 Tiempo de lectura: 26 min

Six years ago I had a system I thought was bulletproof. F5 unders, only on matchups where both starters had K-BB% above 18% and the park was below 100 run factor. The system worked. The ROI was honest. Over 14 months it returned around 6% on staked volume. And in month three of that 14-month run, I went on a six-bet losing streak, was staking 8% of my bankroll on each pick, and watched roughly a third of my season’s capital evaporate inside ten days. The system was right. My staking was wrong. The Kelly Criterion is the formula that fixes that exact mistake.

The Kelly Criterion baseball framework matters in MLB more than in almost any other sport, for one structural reason. Baseball is a high-variance game with narrow edges. Even genuinely sharp systems land in the 1-3% ROI range over a full season. With edges that thin, the gap between staking correctly and staking aggressively is the entire question of whether you finish the season ahead or behind. A profitable system staked at the wrong size can produce a losing year. That is the maths trap most punters never see until it has already cost them money.

This guide walks through what the Kelly Criterion actually is, why full Kelly is structurally wrong for MLB even when the formula technically endorses it, how fractional Kelly (half, quarter and below) trades expected growth for sleep at night, the probability-estimation pitfalls that wreck the whole exercise if you do not respect them, and the UK regulatory context that puts hard ceilings on staking regardless of what the maths recommends. The final section walks through a worked monthly portfolio so you can see the numbers move through a realistic example.

The headline message is simple. The Kelly formula is not advanced. It is the precondition for any system to have a chance to prove itself across a 162-game season. Get the staking wrong and you will never live long enough to find out whether your reads were any good.

The Kelly Formula and What Each Variable Actually Measures

The formula was published by John Kelly Jr. at Bell Labs in 1956. The reason it has survived 70 years of academic and practical scrutiny is that it provably maximises the long-run growth rate of a bankroll exposed to repeated favourable bets. Nothing else gets you to the same growth rate over an infinite horizon. That is a strong mathematical claim and it is the source of the formula’s authority.

The formula itself is f-star equals (bp minus q) divided by b. Translated into English: the optimal fraction of your bankroll to stake on a bet equals your edge divided by the price you are getting. The variables are: f-star is the fraction to stake, b is the net decimal odds minus one (so a price of 2.00 gives b equal to 1.00, a price of 2.20 gives b equal to 1.20), p is your estimated probability of winning, and q is the complement (one minus p).

Let me run through a numeric example. Suppose you are looking at a +120 American odds price, which converts to 2.20 in decimal. Your model says the bet wins 50% of the time. The variables plug in as: b equals 1.20, p equals 0.50, q equals 0.50. The numerator is (1.20 times 0.50) minus 0.50, which equals 0.10. Divide by b: 0.10 divided by 1.20 equals 0.083. So Kelly says stake 8.3% of your bankroll on this bet.

The intuition is that the formula scales the stake by two things in parallel: how big your edge is, and how much the price compensates you per unit risked. An 8.3% recommended stake on a 50/50 bet at 2.20 is enormous. Most experienced bettors would not stake nearly that much on a single MLB game. That gap between what the formula recommends and what a seasoned punter would actually do is the entire reason the next section exists.

Two more worked numbers to anchor the formula. A bet at -110 American (decimal 1.91) with a 55% projected win rate: b is 0.91, p is 0.55, q is 0.45. Numerator equals (0.91 times 0.55) minus 0.45, which is 0.0505. Divide by 0.91: 0.0555, or about 5.6% stake. The same probability at +110 (decimal 2.10): the recommended stake jumps to roughly 14.1%.

That second number, 14.1% of bankroll on a single bet, is where most punters intuitively recoil. The formula is telling you that if you genuinely have a 55% read at +110, the optimal stake is more than one seventh of your entire roll. The maths is correct. The application is where the trouble starts.

Why Full Kelly Quietly Destroys MLB Bankrolls

I have a friend who is a maths teacher. He started betting baseball with full Kelly because the formula is clean and the proof is elegant and he is the kind of person who trusts a proof. Inside three months he was down 40% from peak. The system was profitable on paper. The full-Kelly application was destroying him in practice. Three reasons that show up specifically in MLB.

The first is baseball’s structural variance. Roughly 4 in 9 MLB games (around 44%) are won by the underdog across a full season. That win rate is high enough that almost every favourite you bet is going to lose more often than your read might suggest in the short run. Even with a genuine edge, the variance around the expected outcome is huge. A 10-bet sample at 55% true win rate has a one-in-five chance of going 3-7 or worse. Full Kelly on every one of those bets means catastrophic drawdown on the back of pure variance, not bad reads.

The second is that your probability estimate is never as precise as the formula assumes. Kelly is calibrated on the assumption that p is known exactly. In reality, p is your best guess. If your 55% read is actually 53%, the formula is over-staking by a meaningful amount. The penalty for over-staking is mathematically asymmetric: stake 20% too much and you lose growth, stake 50% too much and you accelerate toward ruin. The formula has zero tolerance for probability misestimation, and you are misestimating to some degree on every bet.

The third is edge size. MLB edges, for genuinely sharp systems, sit in the 1-3% ROI range over a full season. That translates into very small per-bet edges, often in the 1-2 percentage point band between your true probability and the implied probability of the price. Full Kelly on a 1-point edge is already a tiny stake. Full Kelly on a perceived 4-point edge that is actually 1 point is four times the stake the maths recommends, with the corresponding fourfold acceleration of drawdown.

The compounding effect of these three factors is what makes full Kelly catastrophic in baseball specifically. A betting analyst once put the philosophy of systems neatly: I tend to look for systems that win 75 percent or more of the time over at least a five-year period. The pursuit of 75% win-rate systems is a fantasy in baseball, and any application of Kelly that assumes those rates is doomed.

The numerical illustration is sobering. A system with a true 3% ROI staked at full Kelly across 500 bets shows median outcomes around +50% growth, but the 5th percentile of outcomes shows drawdown of 30-40% at some point during the season. Most punters do not have the stomach for that drawdown, and many quit the system before the recovery, locking in the loss. Full Kelly is mathematically optimal for an infinite-horizon, infinite-tolerance bettor. Nobody is that bettor.

Half, Quarter and the Practical Fractional Approach

The industry solution to the full-Kelly problem is fractional Kelly. You take the formula’s recommended stake and multiply it by a fraction less than one. Half Kelly stakes 50% of what the formula recommends. Quarter Kelly stakes 25%. Eighth Kelly stakes 12.5%. The trade-off is between expected long-run growth and short-run drawdown tolerance.

The maths on fractional Kelly is friendly. Half Kelly captures roughly 75% of the long-run growth rate of full Kelly with half the volatility. Quarter Kelly captures around 44% of the growth with a quarter of the volatility. The growth-to-volatility ratio improves as you fraction down, up to a point. Below eighth Kelly the marginal improvement in volatility reduction is smaller than the lost growth, and you are essentially staking flat with a small overlay.

The professional consensus, as far as I can read it from public-facing materials, lands at quarter to half Kelly for sports betting. Quarter Kelly is the comfortable default for most operators of any size. Half Kelly is for bettors with high conviction on their probability estimates and the bankroll cushion to absorb drawdown. Below quarter Kelly is for bettors who do not fully trust their own probability estimates, which describes most of us when we are honest about it.

For MLB specifically, my working practice is quarter Kelly with a hard cap on maximum stake of 2.5% of bankroll. The cap matters because Kelly occasionally produces stake recommendations that look reasonable on paper but feel wrong in the gut. A 6% stake recommendation on a perceived double-digit edge is usually a sign that your probability estimate is too aggressive. Capping the stake forces a sanity check.

The reason quarter Kelly works for baseball: it absorbs probability misestimation gracefully. If your 55% read is actually 53%, full Kelly is over-staking by roughly 70%. Quarter Kelly is over-staking by 17.5% in absolute terms, which is annoying but not destructive. The drawdown profile of quarter Kelly is materially gentler, and the cumulative ROI over a full season is close enough to full Kelly that the trade is worth it.

One refinement that has helped me: variable fractioning by bet type. F5 moneylines and run lines, where my probability estimates are reasonably confident, get quarter Kelly. F5 totals, where the projection is more sensitive to lineup confirmation and weather, get eighth Kelly. Futures and props, where my edge is more speculative, get tenth Kelly or flat staking at 0.5% of bankroll. The fraction tracks the confidence in the underlying probability, not the headline attractiveness of the price.

Probability Estimation: Where the Whole Thing Falls Apart

The single biggest failure mode of Kelly in practice is not the formula. It is the probability estimate that goes into the formula. Every Kelly application I have seen go wrong has gone wrong at the p input, not at the arithmetic.

The reason is psychological. Bettors estimate their probability after they have already decided they like the bet. The estimate is anchored on the decision, not on a neutral assessment of the matchup. If I have spent twenty minutes researching a starter and the conclusion is that I like his price, my probability estimate is going to drift upward to justify the time investment. That is the inflation Kelly cannot survive.

Concretely: if I think my true edge is 4 percentage points (say I have 54% true win rate on a 50/50 implied bet), but my actual edge is 1 percentage point because I have overestimated my read by 3 points, then full Kelly stakes me at four times the maths-correct amount. Even quarter Kelly stakes me at four times what I should be staking on that bet. The error compounds across every bet I make, and the bankroll erosion is not from variance, it is from systematic overestimation of edge.

The defence against this is twofold. The first is closing line value. CLV is the comparison between the price at which you placed your bet and the price at which the market closed. If you consistently bet at better prices than the closing line, the market is telling you that your reads are sharper than the bookmaker thought. If your CLV is consistently negative, the market is telling you the opposite. A positive average CLV over 100 or more bets is the cleanest external evidence that your probability estimates are calibrated rather than inflated. The dedicated article on closing line value walks through how to measure it and what counts as a healthy edge for a UK bettor.

The second defence is a pre-commitment to your probability estimate before you check the price. If you write down your estimate before looking at the line, you cannot anchor your estimate on the price. If you look at the price first and then estimate, your number will drift toward whatever justifies the bet you were already leaning toward. The discipline is small but the effect is structural.

I keep a spreadsheet of every bet I place, with two columns: my probability estimate, and the implied probability of the price I took. After 100 bets, I can compare the average difference to my actual win rate. If my probability estimates average 8 points above implied and my actual win rate is only 3 points above implied, I have a 5-point systematic over-estimation that is destroying my Kelly application. The fix is to scale my estimates down by 5 points across the board.

The harder truth is that most punters never do this. They run Kelly on inflated estimates, watch the bankroll bleed, and conclude the formula does not work. The formula is fine. The input is broken.

UK Context: Affordability Checks, Account Limits and Kelly

The UK context puts a ceiling on Kelly application that does not exist in unregulated markets. Three regulatory features matter to MLB punters: the financial vulnerability check threshold, the structural assumptions baked into UKGC compliance, and the practical cost the industry is absorbing to operate under the new rules.

The first feature is the light-touch financial vulnerability check. It triggers at net deposits of £150 over a rolling 30-day period, and the first statutory levy invoices were issued by the UKGC in September 2025. The trigger is account-level, not bet-level, which means it captures the cumulative pattern of how much money you are moving onto the platform, not how much you are wagering per game. A Kelly punter on a £2,000 bankroll, betting 2% stakes on 30 baseball games a month, is moving £1,200 of stake-volume but the actual net deposit might be much lower depending on results. The check is on deposits, not on stake.

The second feature is that the UKGC estimates around 3% of active gambling accounts will fall under the financial risk assessment regime. That number reflects the targeted nature of the checks, but it also tells you something about who they target. The accounts that get flagged are not the average punter making small bets. They are the accounts with elevated deposit patterns relative to demographic expectations. A Kelly bettor who scales up stakes following a winning streak is exactly the deposit profile most likely to trigger a check.

The third feature is industry cost. EY has estimated the annual cost to the gambling industry of running affordability checks at £125+ million. The reason that matters to you as a punter is that operators are looking for ways to manage their compliance burden, and one of the easier moves is to apply tighter account-level limits on bettors whose patterns trigger the check. Even if you pass the verification, the account may be flagged for closer monitoring, with stake limits applied unilaterally.

How does Kelly interact with this? Three practical adjustments. First, do not scale your Kelly stakes upward purely based on bankroll growth from one or two big winning months. Maintain a deposit pattern that is consistent and modest, even when the bankroll allows for more.

Second, use a quarter Kelly cap that converts to a hard absolute maximum stake in pounds, not just a percentage of bankroll. A fixed absolute cap, say £50 per bet, gives you a known ceiling regardless of how the bankroll is trending.

Third, treat the safer-gambling caps in your account settings as binding, not aspirational. Set the personal deposit limit at a number that reflects your actual budget for losses, not the number Kelly recommends as the upper bound. The personal cap is the failsafe when the bankroll is having a bad month and the temptation to chase reappears.

Building a Hundred-Unit Bankroll Around F5 and Run Line Plays

The model I use for bankroll construction is the hundred-unit framework. Take your total bankroll and divide it into a hundred equal units. Each Kelly stake gets expressed as a number of units rather than a percentage, which keeps the maths intuitive and reduces the temptation to drift the stake size game by game.

For a £1,000 bankroll, each unit is £10. A 2% Kelly stake is 2 units, £20. A 0.5% stake is half a unit, £5. The unit notation also makes it easier to track ROI: if you finish the season up 30 units, that is a 30% return on the initial bankroll, regardless of the absolute pound amount.

Rebalancing is the other part of the bankroll construction question. The textbook advice is that you should recalibrate your unit size as the bankroll grows or shrinks. If your bankroll doubles, your unit doubles, and the absolute stakes scale up. If your bankroll halves, the unit halves, and you stake smaller in absolute terms to protect what remains.

In practice, continuous rebalancing creates two problems. The first is the deposit-pattern issue from the previous section: scaling up the unit as the bankroll grows means depositing or staking more in absolute terms, which can trigger affordability checks unnecessarily. The second is the psychological problem of staking less in absolute terms after a losing month, which feels like surrender even when it is the right thing to do.

The compromise I use is quarterly rebalancing. The unit size is fixed for three months at a time, then recalculated based on the bankroll at the end of the quarter. Within a quarter, the absolute stakes are stable. At the quarter boundary, the unit gets updated. This smooths out the rebalancing problem and avoids the daily temptation to adjust.

The other piece of the construction is separating «hot» and «cold» bankroll. The hot bankroll is the money I am actively betting with: the unit pool, the deposits in the betting accounts, the immediate cash. The cold bankroll is the buffer: money that I have set aside that does not get deployed unless the hot bankroll suffers a drawdown that triggers a top-up. The split is usually 70/30 in favour of the hot bankroll. The 30% cold buffer provides resilience against the long drawdowns that even good systems experience, and it removes the psychological pressure to chase during bad runs because the cold bankroll is the safety net.

Replenishment of the hot bankroll from the cold bankroll happens at the same quarterly cadence as rebalancing, and only if the hot bankroll has drawn down by more than 25% of its starting value. Below that threshold, you ride out the variance. Above it, you top up from the cold pool.

A Worked Monthly Portfolio at Quarter Kelly

Let me run through a realistic monthly portfolio so the numbers are visible. Starting bankroll: £1,000. Unit size: £10. Staking: quarter Kelly with a 2.5% cap, F5 moneylines as the primary product. Average edge: 2%, in the realistic band for a sharp but not exceptional system.

The bet count for a month of F5 betting comes out to roughly 30 bets. Not every game makes the filter; on a 15-game slate you might find 1-2 F5 spots that pass the quality threshold from the pitcher analysis framework. Across a month with around 350 games on the calendar, the filter produces 30-35 bets at the kind of edge worth taking.

Stake sizing at quarter Kelly on a 2% edge bet typically lands in the range of 1.5 to 2.5 units, which is £15 to £25 per bet. The 30 bets at an average of £20 each produces stake volume of £600 in the month. The expected value at 2% ROI is £12 of profit. The standard deviation around that expected value is much larger, in the range of £100 to £150, which is where the variance discussion bites.

What does a realistic month look like? The median outcome is roughly £10 of profit, slightly below the expected value because of the bookmaker margin on the prices. The 25th percentile outcome is around £80 of loss. The 75th percentile outcome is around £100 of profit. The 5th percentile, the bad month, is around £200 of loss, a 20% drawdown of the bankroll. The 95th percentile, the great month, is around £220 of profit.

The distribution tells you something important. Even with a system that has a real 2% edge, half of all months will produce essentially flat or slightly negative results. The variance dominates the signal at the monthly level. The 2% edge only becomes visible after 6 or more months of accumulated bet volume, where the noise washes out and the underlying drift takes over.

Maximum drawdown across a 12-month run of this portfolio is typically 25-35% of starting bankroll. If a 30% drawdown will cause you to abandon the system or change your staking pattern, the system was not set up correctly for your psychological tolerance. The solution is to fraction down further: eighth Kelly produces maximum drawdowns in the 15-20% range, with proportionally lower expected return.

The cumulative ROI at year-end, for a 2% edge system at quarter Kelly, lands at a median of around 9-12% of starting bankroll. That is £90-£120 of profit on a £1,000 starting bank, across 350-400 bets. It is not glamorous, and it is genuinely good. Most punters underestimate how slowly real edges compound and how much variance has to be tolerated to capture them. The Kelly discipline is what lets you finish the year inside the profitable range. Without it, the same system on the same reads can finish anywhere from -25% to +25% on the basis of staking decisions alone.

Common Kelly Questions From UK Punters

The questions punters ask most often when they first encounter the framework.

Stake Discipline as the Precondition for Edge to Show

The Kelly Criterion is not advanced staking. It is the precondition for any system to have a chance of proving itself across a baseball season. Without it, even a profitable system can produce a losing year on the back of pure variance. With it, a profitable system has the structural protection it needs to survive the drawdowns that the sport will inevitably produce.

The discipline involves two things. First, fraction down. Quarter Kelly with an absolute cap is the working default for MLB; full Kelly is reserved for fantasy land. Second, respect the probability estimate as the weakest link. The maths will reward calibrated estimates and punish inflated ones. CLV is the only external check on whether your estimates are honest, and it should be the metric you watch more closely than your monthly P/L.

The UK regulatory environment imposes a hard cap on top of the Kelly recommendation. Affordability checks, deposit thresholds and operator-level scrutiny all push in the direction of more conservative staking patterns than the formula alone would recommend. That is not an inconvenience to be worked around; it is a useful counter-pressure against the punter’s natural tendency to scale up after winning streaks. Treat the UKGC framework as a partner to Kelly, not an obstacle. Both are trying to keep you in the game long enough for the edge to show up.

Preparado por la redacción de «mlb Betting Systems».